A long-standing puzzle in cryptography—how to compute on data while it remains encrypted—may finally have a workable solution. Intel, in collaboration with Microsoft and DARPA, has built a specialized processor that accelerates a technique called fully homomorphic encryption (FHE) by roughly a million times compared to standard software. This leap could move FHE from academic journals into real-world systems. Since its conceptual proof in 2009, FHE has promised a new standard of privacy: a cloud server could analyze medical records or financial transactions without ever decrypting the sensitive information. The catch was speed. The complex mathematics involved could slow processing to a crawl, making it impractical. Intel's new application-specific integrated circuit (ASIC), developed under a DARPA program, is engineered to handle the specific, heavy computations FHE requires, dramatically cutting that delay. The technical report, detailed in IEEE Spectrum, notes the chip tackles the core bottlenecks: massive polynomial multiplications and the movement of huge encrypted data blocks. By integrating large on-chip memory and units designed for these tasks, it reduces a performance gap that once spanned six orders of magnitude to a far narrower margin. This isn't a product you can buy tomorrow. It's a research prototype. The path to commercial data centers involves manufacturing economics, software tool development, and competition from other startups exploring different acceleration methods. Yet, the implications are clear for sectors like healthcare and finance, where privacy regulations tighten each year. The work signals that computing on permanently encrypted data is transitioning from a theoretical ideal to an engineering problem now being solved.

Source: Webpronews